Why do LLM agents return malformed JSON?

Older models (GPT-3.5, Claude 2) frequently drop closing brackets, add trailing commas, or escape quotes wrong. Modern models (GPT-4o, Claude Opus 4, Gemini 2.5) are much better but still occasionally fail on edge cases. Tool-use APIs with structured output guarantees (function calling) solve this at the API level.

What's JSON mode in OpenAI / Claude?

Both vendors now support 'JSON mode' that forces the model's output to be valid parseable JSON. OpenAI calls it response_format: {type: 'json_schema'} with strict schema. Claude has tool-use schema enforcement. These are more reliable than prompting for JSON in the instruction.

Should I repair malformed JSON programmatically?

Yes, for production. Libraries like json-repair (Python) and jsonrepair (JS) fix common LLM output mistakes. Catch-and-repair pattern: try strict parse first, fall back to repair, fall back to re-prompting the LLM with the error. Reduces agent failure rates by 50-80%.

What are common JSON agent mistakes to watch for?

Single quotes instead of double. Python True/False/None instead of JSON true/false/null. Trailing commas. Un-escaped quotes inside strings. Missing closing brackets on nested arrays. Comments (not valid JSON). Our validator catches these with line:col error messages.

AI & Prompt Tools · Free tool

Agent JSON Validator

Paste tool-call or agent-output JSON — parse, pretty-print, highlight errors, and count keys and depth.

Updated June 2026

JSON schemaTool-call output

Overall

All checks passed

+Schema parses as JSON
+Output parses as JSON
+required field "city" present
+required field "units" present
+"city" type = string
+"units" type = string
+"days" type = number

Supports required-field checks and shallow type checks (string, number, boolean, array, object, null). Not a full JSON-Schema validator.

Found this useful?Email Buy Me a Coffee

What it does

Validate JSON emitted by agents and tool calls — with line numbers and key counts.

Embed this tool on your siteShow snippet

Paste this snippet into any page. Loads on-demand (lazy), no tracking scripts, and sized to most dashboards. Replace the height to fit your layout.

<iframe src="https://freetoolarena.com/embed/agent-json-validator" width="100%" height="720" frameborder="0" loading="lazy" title="Agent JSON Validator" style="border:1px solid #e2e8f0;border-radius:12px;max-width:720px;"></iframe>

Embed docs →

How to use it

Paste the JSON.
Read parse status and errors.
Copy the pretty-printed output.

Frequently asked questions

Why do LLM agents return malformed JSON?: Older models (GPT-3.5, Claude 2) frequently drop closing brackets, add trailing commas, or escape quotes wrong. Modern models (GPT-4o, Claude Opus 4, Gemini 2.5) are much better but still occasionally fail on edge cases. Tool-use APIs with structured output guarantees (function calling) solve this at the API level.
What's JSON mode in OpenAI / Claude?: Both vendors now support 'JSON mode' that forces the model's output to be valid parseable JSON. OpenAI calls it response_format: {type: 'json_schema'} with strict schema. Claude has tool-use schema enforcement. These are more reliable than prompting for JSON in the instruction.
Should I repair malformed JSON programmatically?: Yes, for production. Libraries like json-repair (Python) and jsonrepair (JS) fix common LLM output mistakes. Catch-and-repair pattern: try strict parse first, fall back to repair, fall back to re-prompting the LLM with the error. Reduces agent failure rates by 50-80%.
What are common JSON agent mistakes to watch for?: Single quotes instead of double. Python True/False/None instead of JSON true/false/null. Trailing commas. Un-escaped quotes inside strings. Missing closing brackets on nested arrays. Comments (not valid JSON). Our validator catches these with line:col error messages.

Learn more

Explore more ai & prompt tools tools

100% in-browserNo downloadsNo sign-upMalware-freeHow we keep this safe →